@xiaohu: This interview is the most correct and accurate about FDE. After all, Palantir does FDE best. Palantir CEO: Models will become something everyone has; the hard part—and the valuable part—is embedding them into the complex systems of real enterprises. Doing this requires being immersed in customer sites long-term, not sitting in San Francisco...

About FDE, this interview says it most correctly and accurately

After all, Palantir is the best at FDE

Palantir CEO: Models will become something everyone has—stuffing them into the complex systems of real enterprises is what’s hard, and that’s where the real money is.

Doing this requires long-term immersion on-site with clients, not just sitting in San Francisco writing code to get it done.

The engineers those companies hire are extremely good; the problem is they don’t deal with enterprise clients and don’t understand the real business logic or technical challenges.

All day long they’re sipping lattes and reading reports whose technical underpinnings they don’t grasp, then guessing at customer needs to get their work done…

Seedance 2.5 is good, sure it is

But honestly, it’s too expensive to use.

Yesterday I recharged ¥1,200 to test it, and it was gone in no time.

However, I did discover a more powerful use for it…

Seedance 2.5 can actually be used to produce industrial-grade data across various domains: robot training data, autonomous driving data, dangerous experiment simulations, synthetic data for rule-violating behavior, and more.

Here are my test results ↓